D2D Cont'd - The Response to our Nvidia Call
Good luck with that....
The Response So Far
It has been a little over a month since we launched coverage of Nvidia, so we thought it would be useful to revisit our thesis and to reflect on the Street’s response.
When we launched as apparently the only analyst with a Sell rating on the stock, we expected a fair amount of pushback. But that is not what we encountered. The press disagrees with us, but they also welcome controversy. We do not engage with retail investors, but the few who reached out to us also disagree, vehemently. We are aware of a community of retail investors who think we were not strong enough in our sentiment. As noted, we did not respond to or engage with either side. Perhaps the strongest disagreements came from others on the sell-side. We spoke with one former semis analyst, now sitting in a corporate role elsewhere. When we told him about our rating, he paused and replied “You have a sell rating on Nvidia?...Good luck with that.” Probably the most entertaining conversation we had all week.
But where we found almost no disagreement was with investors. We came back to the sell-side, ready for a fair share of contentious discussions, but our views on Nvidia did not go down that path. The only disagreement we heard from the buy-side involved timing. Some argued that we were late to the call – Nvidia had underperformed much of the year already. Others argued that we were too early – the company’s troubles would come after the Blackwell launch when demand for AI dried up or competition increased. These are both valid, worth a debate. That being said, our view sits in between.
Our Thesis
We hold that as the company’s delivery constraints gradually ease through this year we will find the real equilibrium between supply and demand. Yes, there is a huge order book for Blackwell, but this is largely scoped out and priced in, creating a bias towards the downside. Put simply, there are many more things that can surprise to the downside than to the upside. In no particular order our concerns include:
Supply chain bottlenecks. These have been a moving target for several months going back to the board re-design for thermal issues early last year, through CoWoS constraints, now to finally assembly shortages. Along with several other matters. Many of these are resolved, but there ongoing series speaks to a supply chain that is not fully formed.
Tariffs and geo politics.
Construction delays and electricity shortages causing deployment delays.
Ongoing configuration and stand-up issues also potentially causing deployment delays and thus order cuts.
The potential for double and triple orders to unwind. These seem less of a problem now, but these things tend to get discovered at the least convenient time.
Insufficient working capital in the supply chain. To reach full expectations for the Blackwell build-out the ODMs in particular are going to need to carry some significant amounts of inventory. These are big numbers, so they may not have sufficient resources to carry it through. We imagine there is someone out there willing to finance it all, but we can easily see this tripping up deployments somewhere along the line.
All semis deployments face similar deployment issues, but Nvidia’s scale pushes the limits of the industry in a way that creates real risks of order delays and push-outs.
And underneath all of this is the question mark about related party sales. Many of Nvidia’s customers are themselves recipients of equity investments from the company. We are not suggesting anything untoward, but these kinds of arrangements are effectively a form of leverage which benefit the company when things are going well, but will likely magnify downturns when things are not.
But when it comes down to it all the area in which our thesis is most vulnerable is the common refrain we heard. If Nvidia does encounter some form of hiccup in the Blackwell roll-out “The Street will just look through it.” The idea that any short-term supply-chain driven disruption will just get ignored by investors who are focused on the long-term AI story. We heard this one a lot.
The Long-Term Outlook
Which leads to the long-term outlook for the company. We see two axes along which we need to evaluate Nvidia:
The demand for AI
Nvidia’s share of compute satisfying that demand
Of course there is an immense amount of debate about AI right now. On one side there are the true believers that hold Artificial General Intelligence (AGI) is imminent. At the other end of the spectrum are those who say “It’s a Bubble”, bound to deflate, and every variation in between.
There are large swathes of the population who seem convinced AGI is right around the corner. But we have noticed that view is more widely held the further the observer is from actual implementation. True, there are many inside the big AI labs who are believers, but much of the rhetoric around AGI comes from the press and politicians. We know of many long-time AI practitioners who think current LLM models are insufficient to get to true ‘reasoning’ (whatever that means), putting serious questions around the timing (or achievability) of AGI.
Our view is that no new technology as potentially disruptive as AI could be ever gets adopted in a straight line. Absent an AGI surprise, we think adoption of AI takes a pause some time in the next few years to sort out. We readily admit we are resorting to pattern matching with the Internet which needed 2002-2007 to metamorpihize from Pets.com and Webvan into Amazon, Google and Facebook. But we think there are lot of people out there who see a similar pattern.
That being said, we do firmly believe that AI will eventually generate significant productivity enhancements for everyone (until the robots take over). It’s just that the pace of adoption will not be even.
The other big question is the degree to which Nvidia will dominate that growth. Their position looks very strong right now, but it is not unbreachable.
The company currently has little competition, but that is changing. AMD looks set to retain its perennial role as second source. We do not think they are going to overtake Nvidia any time soon, but they are getting close to the point where their presence in the market can provide at least some degree of negotiating leverage, putting a break (but not really a cap) on Nvidia’s pricing.
Nvidia’s biggest end-customers are also working on their own alternatives. Nvidia argues that their core competency rests in designing these systems, while it is definitely not core to those customers, and with time that will give Nvidia a perennial lead. The semiconductor industry has been having this argument – between merchant and internal silicon – for fifty years, like a pendulum swinging from one side to the other. Some, maybe many, of those internal chips will fail and get abandoned, but again their very existence presents a constraint for Nvidia. The math on internal silicon usually pencils out in favor of merchant, but not always, and the higher the merchant provider’s prices the better the math looks for internal.
Inference is not a lock for Nvidia. This is probably the most hotly contested area right now. In theory, the rise of inference workloads present a major opportunity for Nvidia’s competitors to win some share, but in practice that does not seem to be the case yet. Nvidia is doing significant inference business today. We think the jury is still out here, and a lot will depend on what workloads really take off. Right now, Nvidia has an incumbent’s advantage, but the size and scope of inference needs are so large, there are real opportunities for other approaches here.
2 X 2 Matrix
All of which leads to a management consultant’s platonic ideal 2x2 matrix to think about the company over a long time horizon.
Low AI Growth, Low Nvidia share. This is the ultimate bear case for the company. If AI fizzles out, and what little does get done, can be run on anyone’s silicon, then Nvidia declines pretty sharply.
Low AI Growth, High Nvidia share. This is not a terrible outcome, with the company dominating a niche market. We are not going to run the numbers on this, but it seems fairly likely most shareholders today would see that as disappointing.
High AI Growth, Low Nvidia share. We think this would take the form of AI seeing explosive growth to the point that the cloud providers and software companies need to scramble to bring down costs. This would involve moving as much inference to the edge (i.e. devices they do not pay for) as possible, and getting very aggressive about internal solutions. In this scenario, Nvidia would lose share and/or pricing. They would not fade into irrelevance but instead just be a leader among many.
High AI Growth, High Nvidia share. If this happens, and you work in semiconductors, it’s probably best if you have never done anything to earn bad marks from the company, like having a sell rating on their stock.